Case study 01 — Product thinking & strategy
AI Metadata Generator
How I designed and scaled a serverless AI pipeline that turns video into intelligent metadata.
Vision & strategic rationale
Vision
Democratize metadata generation — enabling creators and studios to automatically understand, tag, and enrich their content at scale using AI.
Strategic problem
Streaming platforms like Disney+ rely on metadata to drive discoverability, personalization, compliance, and creative repurposing. But generating that metadata is still manual, fragmented, and non-standardized.
This project reimagines metadata generation as an AI-native pipeline, one that scales storytelling intelligence the way CDNs scale video delivery.
Guiding product philosophy
Empower creativity
The goal wasn't to replace editors or producers — it was to give them a smarter foundation with structure.
Human + machine
Metadata should be readable, queryable, and narratively meaningful for both humans and machines.
Think platform
Every decision was made to support future extensibility — from small uploads to studio workflows.
Product leadership & system design thinking
Separation of concerns
Distinct S3 buckets for uploads and metadata outputs ensure data hygiene and lifecycle control. This architectural decision enables independent scaling, security policies, and retention strategies for each data type.
Progressive product evolution
Orchestration backbone — Step Functions + job tracking
Functional metadata pipeline — Transcribe → AssembleCanonical
Multimodal intelligence — Rekognition + Bedrock Claude Sonnet 4
Iterative learning
Each version validated an architectural pattern before adding complexity — mirroring how world-class product teams derisk AI systems: ship thin verticals, observe, optimize, expand.
Business framing
This architecture could underpin a metadata intelligence layer for Disney+, ESPN, and Marvel Studios — automatically generating scene-level summaries, character tags, content advisories, and recommendation features.